Inflammation during early post-acute COVID-19 is associated with reduced exercise capacity and Long COVID symptoms after 1 year

Participants

LIINC participants with a history of nucleic-acid confirmed SARS-CoV-2 infection who had completed an echocardiogram study visit were offered participation in additional cardiopulmonary testing in order of date of infection (earliest first) irrespective of symptom status. Participants were excluded if they were pregnant or had a history of significant cardiopulmonary disease including congenital heart disease, heart failure, myocardial infarction, or heart surgery. Additionally, those with non-MRI compatible implants or claustrophobia were excluded from CMR; those with estimated GFR <30ml/min/1.73m² were excluded from receiving gadolinium contrast. We excluded those unable to exercise on cycle ergometer from CPET.

Symptoms

Individuals were queried regarding 32 PASC symptoms and using the Patient Health Questionnaire Somatic Symptom Scale.²⁷ At enrollment, participants completed a structured interview about medical history, characteristics of acute infection, cardiopulmonary diagnoses, and symptoms within the previous two weeks. We defined a composite symptom variable for cardiopulmonary PASC including chest pain, dyspnea, or palpitations (“cardiopulmonary symptoms”) in the preceding 2 weeks prior to the study visit and a general PASC composite that included cardiopulmonary symptoms and fatigue (“symptoms”).⁹ Consistent with the WHO definition, all participants had new symptoms without alternative cardiopulmonary explanations and were more than 3 months after SARS-CoV-2 infection.²⁸

Echocardiography

A cardiac sonographer blinded to patient data performed echocardiograms with a GE VIVID E90 machine at the first study visit using a standardized protocol. Post-processing and measurements were performed by a single echocardiographer using GE EchoPAC software according to ASE guidelines,²⁹ as we have previously described.⁹

Blood-Based Measures

Participants had venous blood collected and processed for serum and plasma on the day of the echocardiogram. Samples were batch processed for measurement of high sensitivity troponin I (hs-troponin; ADVIA Centaur® High-Sensitivity Troponin I (TNIH) assay), high sensitivity c-reactive protein (hsCRP; ADVIA® Chemistry CardioPhase™ High Sensitivity C-Reactive Protein assay), and N-terminal prohormone b-type natriuretic protein (NT-pro-BNP; Roche Cobas 6000 Elecsys® proBNP II assay). Most participants had antibodies and additional markers collected at two time points (<90 days and 90-150 days) after symptom onset analyzed by Monogram Biosciences (South San Francisco, CA) using the Quanterix Simoa® platform with Simoa® Assay Kits from Quanterix Corporation (Billerica, MA).³⁰ These included interleukin 6 (IL-6), interleukin 10 (IL-10), glial fibrillary acidic protein (GFAP), neurofilament light chain (NfL), monocyte chemoattractant protein-1 (MCP-1), interferon gamma (IFN-γ), and tumor necrosis factor alpha (TNF-α), and SARS-CoV-2 receptor-binding domain (RBD) immunoglobulin G (IgG). Samples were assayed blinded with respect to patient and clinical information, and assay performance was consistent with the manufacturers’ specifications. Additionally, blood was drawn immediately prior to CMR for measurement of hematocrit for calculation of extracellular volume.

Cardiac Magnetic Resonance Imaging

Multiparametric, sequence-standardized, blinded (technician and reader) cardiac magnetic resonance imaging (CMR) was performed with a 3 Tesla system (Premier, General Electric). The protocol consisted of acquisition of the following sequences after multiplane localizers prior to gadolinium injection: fast imaging employing steady state acquisition (FIESTA) cine in the axial short axis planes, pre-contrast T1 mapping sequences using the MOLLI 5-(3)-3 technique, as well as pre-contrast T2 mapping at the basal, mid, and apical short axis planes and T2 fat-saturated weighted black blood spin-echo images in the short axis plane. 8-10 minutes after intravenous gadolinium injection, phase sensitive inversion recovery (PSIR) late gadolinium enhancement imaging in short axis full stack, 4 chamber full stack, three slices of 2 chamber images, and post-contrast T1 mapping sequences were obtained. Inversion times were individualized to null the myocardium. Measurements were performed by a single reader at a dedicated workstation using Medis (Leiden, Netherlands) and AI-assisted Arterys (San Francisco, CA) under supervision of a senior cardiac imager, both blinded to all clinical variables, and in accordance with Society for Cardiovascular Magnetic Resonance task force recommendations. Arterys software was used for T1 and T2 mapping and ECV calculation using pre- and post-contrast MOLLI sequences.

Cardiopulmonary Exercise Testing

CPETs were performed by an experienced exercise physiologist and noninvasive cardiology nurse practitioner blinded to participant data according to standard protocol using a metabolic cart (Medical Graphics Corporation Ultima CardiO₂) and supine cycle ergometer (Lode Corival CPET) with continuous 12 lead ECG monitoring (GE CASE) and noninvasive blood pressure and pulse oximetry measurement. First, baseline ECG, blood pressure, and rest spirometry including maximum voluntary ventilation (MVV) were measured. We determined the work increase per 1 minute step based on the expected peak VO₂ from the MVV for a goal 10 minute test, rounded to 5 Watts/min increments based on reported exercise (range 10-30W/min) in accordance with guidelines.³¹ Participants underwent a 2-minute rest phase, 2-minute no resistance warm up, and then 1-minute steps. Breath-by-breath oxygen consumption (VO₂) and carbon dioxide production (VCO₂) were measured continuously. Participants were blinded to time, wattage, and peak VO₂ during the test and encouraged to maintain a cadence of ~60 cycles per minute and exercise to their maximum ability, with the test stopped prematurely for severe hypertension, relative hypotension, moderate to severe angina, ventricular tachycardia or couplets, or ischemic changes. Reason for stopping and exertional symptoms were recorded. Exercise effort was assessed by Borg Scale and respiratory exchange ratio (RER). Anaerobic threshold was determined manually by the exercise physiologist using the slope method. CPETs were interpreted separately by two independent readers and differences in interpretation were discussed and resolved through consensus.

We evaluated measured peak VO₂ (in ml/kg/min) and used the Wasserman equations to estimate the percent predicted VO₂ for each participant.^32–34 We classified exercise capacity <85% predicted as reduced. We considered ventilatory limitation if rest spirometry was abnormal, tidal volume did not double over baseline reaching at least 50% of FVC, breathing reserve was <30%, dead space ratio (V_D/V_T) was >0.2, or desaturation occurred. We classified participants as having a cardiac limitation if there were ischemic ECG changes, the oxygen pulse was reduced and plateaued early, AT occurred at less than 40% of predicted maximal VO₂, and if ventilatory efficiency slope (VE/VCO₂) was <32. We defined an abnormal chronotropic response as <80% of the adjusted heart rate reserve (AHRR) calculated as (HR_peak-HR_rest)/(220-age-HR_rest).³⁵ We considered exercise limitation to be likely due to deconditioning, obesity or impaired gas exchange if exercise capacity was reduced without cardiac or ventilatory limitation.

Ambulatory Rhythm Monitoring

After CMR, an ambulatory rhythm monitor (Carnation Ambulatory Monitor, BardyDx) was placed on the chest. Participants were instructed wear it for up to 2 weeks, press the button for cardiopulmonary symptoms, and record symptoms in a diary. Monitors were processed according to BardyDx standard procedures and reports were overread by a cardiologist.

Statistical Design

Adjusted models included potential confounders age, sex, time since SARS-CoV-2 infection, hospitalization for acute SARS-CoV-2 infection, and body mass index. We used logistic regression to estimate associations of parameters with symptoms and linear regression to estimate adjusted mean differences between those parameters among participants with and without symptoms. For count outcomes (button pushes on rhythm monitoring), we used Poisson regression. Non-normally distributed variables (hsCRP, T1 mapping times, premature atrial contraction burden, for example) were log-transformed and findings are reported per doubling or per 10-fold change depending on the range of values. For biomarker data we report unadjusted Pearson’s rho correlation coefficients, adjusted linear regression models, and for longitudinal biomarker data we used mixed effects models with a random intercept per patient. We conducted sensitivity analyses considering other definitions of symptoms and additionally adjusting for potentially relevant past medical history and baseline echocardiographic and spirometry parameters.

The study was approved by the UCSF Institutional Review Board (IRB 20-33000) and all participants provided written informed consent. REDCap was used for data entry. Statistical analyses were performed using STATA version 17.1.

Participant Characteristics

Among 120 participants who completed Visit 1, 46 participants had at least one advanced test performed. The median age was 52 (IQR 42-61), 18 (39%) were female, and 6 (13%) were hospitalized (Table 1). Two notable differences between the subset who underwent advanced testing were earlier date of infection (p<0.001) and a lower proportion with HIV (p=0.06). As most individuals were infected between March and September 2020, no participants had been vaccinated at the time of initial infection, but all had received at least one SARS-CoV-2 vaccine prior to advanced testing.

Symptoms Persist at 18 Months

At visit 1 (median 6 months after SARS-CoV-2 infection), 73/120 (61%) reported PASC symptoms and 56/120 (47%) reported cardiopulmonary symptoms (chest pain, dyspnea, or palpitations). At visit 2 (median 17.0 months), 25/47 (53%) reported PASC symptoms and 19/47 (40%) reported cardiopulmonary symptoms. Among the 25 individuals with symptoms at visit 2, 20 had reported symptoms at visit 1, and 5 developed new symptoms between visits. Nine people reported resolution of symptoms by visit 2. Individual symptom patterns were similar (Supplemental Table 1). Reporting reduced exercise capacity was associated with reporting other cardiopulmonary symptoms (OR 8.04, 95%CI 1.94-33.3; p=0.004); 78% versus 32% with preserved or improved exercise capacity had cardiopulmonary symptoms (p=0.003).

CPETs were Maximal Tests

Out of 40 participants who attended a CPET visit, 39 completed CPET at a median 17.5 months (IQR 15.5-18.4) after SARS-CoV-2 infection (one excluded due to hypertensive urgency). Two CPETs (one with and one without symptoms) were stopped early for hypertensive response after predicted heart rate and exercise capacity were reached but before symptomatic limitation. All other tests were completed to symptom-limited maximal exertion. All tests had a respiratory exchange ratio (RER) >1.05 with no difference in RER achieved by symptom status. No participants were taking beta blockers, non-dihydropyridine calcium channel blockers, or ivabradine at the time of CPET.

Reduced Exercise Capacity on CPET

Overall, 15/39 (38%) participants had reduced exercise capacity (peak VO₂ <85% predicted). Among those with PASC symptoms, 13/23 (57%) had reduced exercise capacity compared to 2/16 (13%) without symptoms (p=0.008 by Fisher’s exact test). The odds of reduced exercise capacity were 3.0 times higher for cardiopulmonary symptoms (95%CI 0.8-11.4; p=0.11) and 9.1 times higher for PASC symptoms (95%CI 1.7 to 49.6; p=0.01). Objectively reduced exercise capacity on CPET was associated with 25 times higher odds of PASC symptoms compared to preserved exercise capacity (95%CI 2.1 to 303; p=0.01).

As shown in Figure 1, peak VO₂ was 21.2±8.2 ml/kg/min among those with symptoms and 28.8±7.7 among those without symptoms, a difference of −7.6 (95%CI −12.9 to −2.3; p=0.006) or 89% versus 111% percent predicted (−22.1% predicted, 95%CI −36.5 to −7.7; p=0.003). The adjusted difference in peak VO₂ was −5.9 ml/kg/min (95%CI −9.6 to −2.3; p=0.002) or −21% predicted (95%CI −35 to −7; p=0.006). Results were unchanged in sensitivity analyses accounting for history of asthma/COPD and hypertension, resting spirometry (FVC, FEV₁, MVV), and baseline echocardiogram (LVEF, LV diastolic function, LV strain, RV strain, and TAPSE) with point estimates ranging from −4.4 to −7.9 ml/kg/min and no confidence intervals crossing 0. The difference in peak VO₂ was smaller among those reporting only cardiopulmonary symptoms: 2.7 ml/kg/min lower (95%CI −6.9 to 1.5; p=0.20; Supplemental Table 2).

• Download figure
• Open in new tab

Figure 1. Exercise Capacity by PASC Symptoms and by HR Response to Exercise (n=39)

Figure 1 Legend: On the top row are box and whisker plots of peak oxygen consumption (VO₂ in ml/kg/min on the left and percent of predicted on the right) among those without (blue) and with chest pain, dyspnea, palpitations, or fatigue (pink). Peak VO₂<85% was associated with PASC symptoms (OR 25.0, 95%CI 2.1 to 303; p=0.01). Mean peak VO₂ was 21.2±8.2 ml/kg/min among those with symptoms compared to 28.8±7.7 ml/kg/min among those without symptoms, a difference of −7.6 ml/kg/min (95%CI −12.9 to −2.3; p=0.006) or 89% vs 111% percent predicted (difference −22.1, 95%CI −36.5 to −7.7; p=0.003). After adjustment for age, sex, hospitalization for acute COVID-19, BMI category, and months since SARS-CoV-2 infection, peak VO₂ was 5.9 ml/kg/min lower among those with versus without symptoms (95%CI −9.6 to −2.3; p=0.002) or −21% (95%CI −35 to −7; p=0.006). On the bottom row are peak VO₂ (ml/kg/min on the left and percent predicted on the right) by reaching an AHRR >80% (normal chronotropic response to exercise) or <80% (blunted chronotropic response to exercise). Not reaching adjusted heart rate reserve >80% during CPET was associated with 15.6x higher odds of having reduced exercise capacity <85% predicted (95%CI 1.30-187; p=0.03).

As shown in Table 2, the work completed was lower and perceived effort was higher among those with symptoms, even though the respiratory exchange ratio (an objective measure of reaching a maximal test) was similar. There were no significant differences on rest echocardiography (Supplemental Table 3) or spirometry (Supplemental Table 4), and most CPET parameters were not significantly associated with symptoms (Table 2). Reaching anaerobic threshold at a lower VO₂, but not as a percent of predicted peak VO₂, was associated with symptoms.

Classification of Reduced Exercise Capacity by Pattern of CPET Findings

A higher dead space ratio at peak exertion was associated with symptoms, but no participants had a ventilatory limitation. One participant had a cardiac limitation (reduced oxygen pulse pressure and ischemic ECG changes), and one had a hypertensive response. One participant’s peak VO₂ was slightly below 85% predicted with no other abnormalities (which could be deconditioning), and three participants had findings most consistent with deconditioning/obesity. Among participants with reduced exercise capacity (<85% predicted), 9/15 (60%) had chronotropic incompetence with an adjusted heart rate reserve (AHRR) <80% and peak heart rate <85% age predicted maximum. One participant’s CPET pattern was most consistent with obesity/deconditioning, but AHRR=80%; we did not classify that participant as having chronotropic incompetence.

Chronotropic Incompetence on CPET

Nineteen participants had AHRR<80%, which was associated with reduced exercise capacity (OR 15.6, 95%CI 1.30-187; p=0.03) and 4.9 ml/kg/min lower peak VO₂ (95%CI CI −9.3 to −0.5; p=0.03). Those with chronotropic incompetence had much lower peak heart rate (117±10 bpm) compared to those with normal peak VO₂ and HR (168±13), an adjusted difference of 48 bpm (95%CI 39 to 56; p<0.001; Supplemental Figure 1). Chronotropic incompetence was associated with 6.4 ml/kg/min lower peak VO₂ (95%CI −11.3 to −1.6; p=0.01), a large and clinically meaningful difference. Compared to individuals with normal peak VO₂ and HR, chronotropic incompetence was associated with reduced heart rate recovery at 1 minute after cessation of exercise (adjusted difference −8.7, 95%CI −15.6 to −1.7; p=0.02). Although achieving adequate heart rate can be a measure of maximal effort, the respiratory exchange ratio was not lower (difference 0.04, 95%CI −0.02 to 12; p=0.19) and the Borg effort was the same (difference −0.2, 95%CI −1.8 to 1.5; p=0.85) among those with chronotropic incompetence.

• Download figure
• Open in new tab

Supplemental Figure 1. Heart Rate during Exercise by Chronotropic Response to Exercise

Supplemental Figure 1 Legend: Mean heart rate is plotted as a function of exercise time normalized to percent of predicted peak VO2: in purple are those with normal exercise capacity (peak VO2 >85% predicted and normal heart rate response (n=16), in teal are those with normal exercise capacity (peak VO2>85%; n=8) and blunted heart rate response (AHRR<80%; n=8), and in yellow are those with chronotropic incompetence (n=9), as described in Supplemental Table 8.

Normal Cardiac Structure and Function on CMR

One participant was not administered gadolinium due to eGFR<30, and 3 participants could not complete CMR due to claustrophobia. CMR demonstrated normal LV and RV volumes and ejection fraction in all participants (n=39, Table 4). Higher extracellular volume was associated with symptoms, but no participants had late gadolinium enhancement suggestive of myocardial scar, and other markers suggestive of cardiac inflammation including native T1 and T2 parametric mapping values were not associated with symptoms. A high proportion of participants (11/39, 28%) had trace or small pericardial effusions with no difference by symptom status. Smaller RV end diastolic volume index and smaller stroke volume were associated with cardiopulmonary symptoms (Supplemental Table 6). Those with chronotropic incompetence had lower stroke volume at rest (−19ml, 95%CI −34 to −3; p=0.02) and higher extracellular volume (6.2%, 95%CI 0.7 to 11.7; p=0.03).

No Arrythmias on Ambulatory Rhythm Monitoring

There were no clinically significant arrhythmias including atrial fibrillation or atrial flutter on ambulatory rhythm monitoring (Supplemental Table 7). There were no statistically significant differences in premature atrial contractions (PAC), premature ventricular contractions (PVC), sinus tachycardia, or supraventricular tachycardias (SVT) by symptom status. Most participants had no PVCs, so we could not exclude meaningful increases in PVC burden given wide confidence intervals. Symptomatic individuals pressed the button 3.3 times more often (95%CI 2.3-4.8; p<0.001). Most button pushes were during sinus rhythm, sinus tachycardia, or supraventricular ectopy (Supplemental Figure 2), except in one participant with PASC in whom button pushes correlated with brief episodes of SVT. Results were similar in sensitivity analyses adjusting for echocardiographic parameters including LVEF, LV strain, and LA volume index.

• Download figure
• Open in new tab

Supplemental Figure 2. Cardiac Rhythms During Button Pushes (n=37)

Supplemental Figure 1 Legend: Individual participant analysis of number of button pushes and assocaited rhythm with each participant on the x axis and the number of times that individual pushed the symptom button on the y axis with each identified rhythm coded by color. Most button pushes were associated with sinus rhythm or sinus tachycardia, with supraventricular ectopy (premature atrial contractions) present among a few individuals especially the 3 with the most button pushes.

• Download figure
• Open in new tab

Figure 2. Correlations between Peak VO₂ and Previously Measured Biomarkers

Figure 2 Legend. Scatterplots and linear trend lines of peak VO₂ by natural log of biomarkers including hsCRP (n=36, median 6 months after infection), hs-Troponin I (n=36, median 6 months after infection), NT-Pro-BNP (n=30, median 6 months after infection), IL-6 (n=33, median 3.5 months after infection), TNF-α (n=33, median 3.5 months after infection), and SARS-CoV-2 Receptor Binding Domain IgG Antibody Level (n=32, median 3.5 months after infection) with unadjusted Pearson’s rho correlations and p-values listed. On unadjusted analysis, hsCRP, IL-6, TNF-α and SARS-CoV-2 antibody levels were highly correlated with peak VO₂, but not hs-Troponin I or NT-pro-BNP. After adjustment for age, sex, BMI, and time of testing, per doubling of biomarker level: peak VO₂ was −2.1 ml/kg/min for hsCRP (95%CI −3.4 to −0.8; p=0.003), −6.4ml/kg/min for TNF-alpha (95%CI −12.3 to −0.4; p=0.04), −2.2 ml/kg/min for IL-6 (95%CI −4.7 to 0.4; p=0.10), and −1.1 ml/kg/min for antibody levels (95%CI −2.5 to 0.3; p=0.13).

Ambulatory Rhythm Monitoring Correlates of Chronotropic Incompetence on CPET

Peak heart rate during CPET correlated with maximum sinus heart rate during ambulatory rhythm monitoring (Pearson’s r=0.73; p<0.001), with an adjusted difference between those with and without chronotropic incompetence of 24 bpm (95%CI 6-41; p=0.01). On ambulatory rhythm monitoring, those with chronotropic incompetence during CPET had a higher average heart rate, higher minimum heart rate, lower maximum heart rate, and lower heart rate variability (Supplemental Table 8). PR intervals were not significantly longer among those with chronotropic incompetence (174ms vs 162ms, +12ms, 95%CI −4 to +27; p=0.13), and no individuals had 2^nd or 3^rd degree heart block.

Markers of Inflammation, Not Cardiac Biomarkers, Early in PASC Trajectory are Associated with Reduced Exercise Capacity and Pericardial Effusions More than 1 Year Later

As shown in Figure 2, markers of inflammation in the blood (hsCRP, IL-6, TNF-α) and SARS-CoV-2 RBD IgG level measured at 3-9 months after infection are highly negatively correlated with peak VO₂ more than one year after infection. Cardiac biomarkers, hs-troponin I and NT-pro-BNP, were not associated with peak VO₂. After adjustment, peak VO₂ was 1.8 ml/kg/min lower per doubling of hsCRP (95%CI −3.3 to −0.2; p=0.03) and 6.4 ml/kg/min lower per doubling of TNF-α (95%CI −12.3 to −0.4; 95%CI p=0.04) with similar but non-statistically significant differences for IL-6 (−2.2, 95%CI −4.7 to 0.4; p=0.10) and SARS-CoV-2 IgG (−1.1, 95%CI −2.4 to 0.3; p=0.12).

The odds of pericardial effusion on CMR were 1.8 times higher per doubling of hsCRP (95%CI 1.09-2.92; p=0.02). Adjusted results for associations between pericardial effusions and IL-6, TNF-α and SARS-CoV-2 RBD IgG at the later time point were not statistically significant, although the confidence intervals do not exclude an effect. There were no significant associations between ECV and inflammatory biomarkers.

Longitudinal serum biomarkers of inflammation, neurologic injury, and SARS-CoV-2 RBD IgG were measured at <90 days from SARS-CoV-2 acute infection (median 52 days from symptom onset) and between 90-150 days (median 124 days from symptom onset) in 34 participants who underwent CPET (Figure 3). Among those with reduced compared to preserved exercise capacity, SARS-CoV-2 IgG (3.00-fold higher mean ratio, 95%CI 1.31-6.88; p=0.009) and TNF-α (1.42-fold higher mean ratio, 95%CI 1.16-1.73; p=0.001) were significantly higher at the early time point (<90 days) with similar patterns in MCP-1 (1.26-fold higher mean ratio, 95%CI 0.99-1.60; p=0.06) and IL-10 (1.23-fold higher mean ratio, 95%CI 0.92-1.65; p=0.17) that were not statistically significant. At the second time point (90-150 days), IL-6 (1.39-fold higher mean ratio, 95%CI 0.93-2.07; p=0.11), (1.19-fold higher mean ratio, 95%CI 0.99-1.43; p=0.06) and SARS-CoV-2 IgG (2.03-fold higher mean ratio, 95%CI 0.90-4.55; p=0.09) were elevated although confidence intervals cross 1 when exercise capacity is dichotomized. All biomarkers decreased over time regardless of eventual exercise capacity, except for IL-6, which increased among those with reduced exercise capacity. Among the neurologic markers, GFAP was lower at the second time point (0.72-fold higher mean ratio, 95%CI 0.51-1.00; p=0.049) and NfL was not different by group at either time point.

• Download figure
• Open in new tab

Figure 3. Change in Biomarkers During the Early Post-Infection Period among those with Reduced versus Perserved Exercise Capacity at ~17 months After Infection (n=32)

Figure 3 Legend: Mean ± standard error for serum biomarkers measured at <90 days from SARS-CoV-2 acute infection (median 52 days from symptom onset) and between 90-150 days (median 124 days from symptom onset) in 32 participants who underwent CPET. Biomarkers include interleukin 6 (Il-6), glial fibrillary acidic protein (GFAP), neurofilament light chain (NFL), monocyte chemoattractant protein-1, interferon gamma [IFN-γ], Interleukin 10 (IL-10), tumor necrosis factor alpha [TNF-α], and SARS-CoV-2 receptor-binding domain (RBD) immunoglobulin G (IgG). Markers of inflammation decreased over time in both groups, except for IL-6, which increased among those with reduced exercise capacity. MCP-1, TNF-α, and IgG were higher at the early time point among those with reduced exercise capacity. TNF-α and IgG remained significantly elevated, and GFAP became significantly lower at the second time point among those with reduced exercise capacity.

Connections between Inflammation, Reduced Exercise Capacity, and Autonomic Responses

One other group previously demonstrated that hsCRP, IL-6, and TNF-α are associated with reduced peak VO₂ three months after COVID-19 hospitalization.³⁶ Our findings suggest that higher levels of SARS-CoV-2 antibodies and inflammation during the early post-acute period are associated with lower peak VO₂ more than a year later. Elevated inflammatory markers, especially hsCRP, IL-6, and TNF-α could be due to higher severity of acute illness,³⁷ persistent immune activation,³⁸ and/or ongoing antigen stimulation from viral persistence.^39–43

The correlation between these markers and peak VO₂ could arise from a common cause of PASC, or these markers could be on the causal path from SARS-CoV-2 infection to symptoms and reduced exercise capacity in Long COVID (Figure 4). Multiple studies have demonstrated that IL-6 may be elevated in PASC.^{30, 38, 44, 45} Intriguingly, in our study only IL-6 did not decrease among individuals who eventually had reduced exercise capacity. Apart from COVID-19, IL-6 impairs chronotropic responses to autonomic signaling in mice⁴⁶ and may regulate energy allocation during exercise.⁴⁷ IL-6 and TNF-α also impair endothelial function in animal models via increasing oxidative stress and suppressing endothelial nitric oxide synthase pathways.⁴⁸ Endothelial and coronary microvascular dysfunction are associated with chronotropic incompetence in the general population^49–51 and have been demonstrated in early PASC.^52–54 Endothelial dysfunction may be associated with reduced VO₂ in PASC;⁵⁵ this could be due to endothelial injury from direct viral infection, or due to effects of persistent inflammation and immune activation.⁵⁶ Elevated inflammation resulting in changes in autonomic function in PASC could explain reduced exercise capacity, chronotropic incompetence, blunted heart rate recovery, reduced heart rate variability and CPET findings others have noted in PASC including altered peripheral oxygen extraction,⁵⁷ preload failure⁵⁸ or inadequate stroke volume augmentation,^{23, 24} and disordered breathing.^58–60

• Download figure
• Open in new tab

Figure 4. Proposed Mechanisms of Reduced Exercise Capacity and Cardiopulmonary Symptoms in Post-Acute COVID-19

Figure 4 Legend: Figure made with biorender.com. We found that higher inflammatory markers (hsCRP, MCP-1, TNF-α) and SARS-CoV-2 receptor binding domain IgG antibody levels measured within the first six months after infection are associated with reduced exercise capacity measured on cardiopulmonary exercise testing more than 1 year after infection. We propose that persistent immune activation and systemic inflammation may cause a dysregulated response to autonomic signaling that many manifest as chronotropic incompetence and cause reduced exercise capacity and symptoms of Long COVID.

Autonomic Function, Sinus Node Function, and Inflammation in PASC

Altered autonomic function is the most likely unifying explanation for the constellation of findings we observed. SARS-CoV-2 may lead to damage of the peripheral autonomic nerve fibers as sequalae of direct viral infection, secondary inflammation of the nerves, autoimmune neuropathy, or alterations in central nervous system regulation. Skin biopsies suggest small fiber neuropathy among those with COVID-19 associated postural-orthostatic tachycardia syndrome (POTS) and abnormal autonomic testing.^{61, 62} Similar to preload failure pathophysiology in POTS, autonomic reflexes govern regulation of preload during exercise.⁶³ Autonomic dysfunction could also occur via infection of autonomic regulatory regions in the brainstem,⁶⁴ or through changes in the amygdala evident on longitudinal pre- and post-infection brain MRIs.⁶⁵ We have previously demonstrated abnormal markers of neurologic injury in PASC in serum⁴⁴ and cerebrospinal fluid.⁶⁶

Alternatively, sinus node remodeling has been hypothesized to reduce sinus node reserve in heart failure,⁶⁷ and SARS-CoV-2 may directly affect sinus node function. SARS-CoV-2 can infect sinoatrial node cells in hamsters and in vitro sinoatrial-like pacemaker cells resulting in sinoatrial node-dysfunction including changes in calcium handling, activated inflammatory pathways, and induced ferroptosis.⁶⁸ Although we did not find evidence of fibrosis on CMR or evidence of sinus node dysfunction on ambulatory rhythm monitoring or during CPET (i.e. sinus pauses, Wenckebach, blocked PACs), sinus node dysfunction could explain chronotropic incompetence.

Inflammation also modifies autonomic and chronotropic responses. Young adults recovering from SARS-CoV-2 have elevated sympathetic activation at rest compared to healthy controls.⁶⁹ Chronic inflammation in other conditions including obesity and heart failure is associated with imbalance between parasympathetic and sympathetic activation manifest in changes in heart rate variability, reduced exercise capacity and chronotropic incompetence.^{49, 70–72} In heart failure, chronotropic incompetence during exercise is negatively correlated with copeptin (C-terminal portion of arginine) and norepinephrine levels.⁷³ FDG-PET studies suggest vascular inflammation⁷⁴ and cardiac inflammation⁷⁵ may be present in early PASC, consistent with findings of elevated inflammatory markers. Systemic inflammation in the setting of stress (as evidenced in the association between amygdalar activity and bone-marrow/arterial inflammation)⁷⁶ has been implicated in the pathophysiology of health disparities^{77, 78} and post-traumatic stress disorder.⁷⁹ Thus, inflammatory responses to direct viral infection, chronic immune activation, localized cardiovascular inflammation, or from the distressing nature of persistent symptoms could alter exertional heart rate augmentation in PASC even without damage to the autonomic nervous system or the sinus node.

Other Studies of PASC using CPET

Our study is consistent with six studies that have reported lower peak VO₂ among those with PASC compared to recovered individuals at 3-6 months after COVID-19.^19–24 Our findings build upon earlier studies by (1) demonstrating reduced peak VO₂ and chronotropic incompetence much later after infection than other published studies, (2) also including evaluation of cardiac inflammation, structural heart disease and cardiovascular arrhythmias, (3) adjusting for likely confounders in our models, (4) including recovered persons as comparators, and (5) and demonstrating associations with higher inflammatory markers early in PASC.

These studies, together with studies that included SARS-CoV-2 uninfected controls and case-studies without controls, have not reached consistent conclusions regarding patterns of reduced exercise capacity. Inconsistency in categorizing exercise-limitation patterns may arise because of selection bias and confounding, the presence of multiple contributing mechanisms, and use of different interpretation algorithms, CPET protocols and adjunctive testing. Deconditioning is commonly reported, perhaps because most studies predominantly included hospitalized individuals and even non-hospitalized individuals experience deconditioning after illness.^{22, 80–84} Although we observed findings consistent with deconditioning or obesity in a few individuals (earlier anaerobic threshold, accelerated heart rate response), our findings argue against deconditioning as the primary explanation. Identifying deconditioning as a cause is challenging because it is also an effect of reduced habitual exercise among symptomatic individuals.

Four other studies have also found that chronotropic incompetence contributes to exercise limitations in PASC.^{23, 85–87} Similarly, “dysautonomia” defined by heart rate parameters was associated with lower peak VO₂ in PASC.⁸⁷ Chronotropic incompetence is not a universal finding in PASC; some exercise protocols make ascertainment of chronotropic incompetence difficult, sensitivity and specificity vary with exercise modality, and including sub-maximal tests or patients on beta-blockers reduces specificity. Nonetheless, recognition of chronotropic incompetence provides insight into the pathophysiology of PASC. Diagnosis of chronotropic incompetence may have prognostic implications: chronotropic incompetence is associated with incident cardiovascular disease, sudden death, and all-cause mortality among men without known coronary artery disease.^88–90

Impaired peripheral oxygen extraction (assessed using invasive CPET) may also contribute to exercise limitations in PASC,⁵⁷ perhaps via changes in autonomic regulation of microcirculatory function.⁹¹ We did not find differences in VO₂/work slope, a noninvasive correlate of measured oxygen extraction. Although more common in PASC than in recovered individuals, ventilatory limitation is uncommon²¹ and only one study has reported ventilatory inefficiency or ventilation/perfusion mismatch from pulmonary vasculopathy.²⁰ Although not observed among any of our participants, dysfunctional (rapid, erratic) breathing or exercise hyperventilation may contribute to dyspnea in PASC as a manifestation of autonomic dysfunction.^58–60

CMR and Ambulatory Rhythm Monitoring Findings

Multiple studies have found CMR findings suggestive of myocarditis without cardiac dysfunction in the early post-acute period.^{15, 92, 93} Consistent with other studies at later time points,^{16, 19} our study reassuringly did not find evidence of abnormal cardiac function or late gadolinium enhancement suggestive of scar. We found higher ECV among those with symptoms, but not significant differences in native parametric mapping times nor associations with inflammatory markers early in PASC, and the clinical meaning of this isolated finding is uncertain.

Our findings are consistent with two studies that did not find significant arrhythmias in early PASC and had inconsistent results regarding PVCs.^{25, 26} A high prevalence of inappropriate sinus tachycardia has been reported in early PASC;⁹⁴ we only identified one individual (without cardiopulmonary symptoms) with resting heart rate >100 bpm and average sinus rate >90 bpm on ambulatory monitoring. Therefore, arrhythmias or inappropriate sinus tachycardia are unlikely to explain symptoms among most individuals with PASC.

Implications for Therapy

Investigation into mechanisms of PASC may benefit from proof-of-concept approaches to identify potential targets for intervention. Given associations with elevated inflammatory markers, targeting inflammatory pathways is worthy of investigation, although whether such interventions can improve exercise capacity, restore autonomic nervous system function, or improve symptoms remains unknown and needs to be tested rigorously. In those with chronotropic incompetence separate from COVID-19 without cardiac implantable devices, exercise is the only intervention demonstrated to improve chronotropic incompetence and improve peak VO₂.⁹⁵ In heart failure, exercise also improves surrogates of autonomic function including heart rate recovery and heart rate variability.^{96, 97} Exercise is an effective treatment for POTS, which may also be related to autonomic responses to stress and has been observed in PASC.^{98, 99}

Limitations

The main limitations of this observational study arise from the non-probabilistic sampling strategy, which is prone to selection bias, and the cross-sectional nature of cardiac measures. The first challenge is the classification of PASC—the statistical significance and magnitude of the difference in peak VO₂ was sensitive to how we defined PASC, but our definition is consistent with current consensus definitions.²⁸ Although participants were not selected based on symptoms, volunteer bias may overestimate the prevalence of reduced exercise capacity and possibly the magnitude of the difference, but should not affect classification of limitation patterns or identification of chronotropic incompetence. We did not include an uninfected comparator group, which could have strengthened our ability to make inferences regardless of symptom definition. As others using CPET have noted, selecting an appropriate control group can be challenging. We do not have pre-infection exercise tests and prior fitness is associated with post-infection peak VO₂; if fitness is associated with risk of PASC, it would be an unmeasured confounder. Although we adjusted for important measured confounders and conducted sensitivity analyses adjusting for potential confounders, there are likely residual confounders (including unmeasured ones like pre-COVID fitness). In terms of measurement, we did not perform invasive CPET, stress echocardiography, stress CMR, or stress ventriculography. Lastly, we did not have contemporaneous biomarker data with CPETs to ascertain whether a hit-and-run transient inflammatory process or ongoing inflammation is more likely.

Conclusions

In conclusion, we found that reduced exercise capacity on CPET is associated with symptoms of PASC at 18 months following SARS-CoV-2 infection. Reduced exercise capacity in PASC is associated with chronotropic incompetence and higher levels of inflammatory markers and antibody levels in the early post-acute period. While a significant proportion had pericardial effusions (also associated with hsCRP) and increased extracellular volume fraction was associated with symptoms, we did not find strong evidence of prior or ongoing myocarditis. Further investigation into mechanisms of cardiopulmonary PASC should include evaluation of inflammatory pathways, chronotropic function, and the autonomic nervous system to identify potential therapeutic targets.